Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter.
Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.
OCR for page 115
CHAPTER 4
ANALYSIS OF FERTILITY DETERMINANTS AT THE NATIONAL LEVEL
This chapter uses multivariate regression analysis to
explain how changes in average parity among married women
in Brazil relate to their changing socioeconomic charac-
teristi~s during the accelerated fertility decline of the
early 1970s. In so doing, it attempts to provide a more
systematic assessment of the hypotheses presented in the
last two chapters, supported by tabular evidence, about
socioeconomic differences in fertility decline in Brazil
and the forces behind those dif ferences.
The decision to focus on variability in the average
parity of married women was based on several considera-
tions. First, fertility was measured according to average
par ity rather than the last birth reported because of the
questionable reliability and comparability of the latter
in studying changes over time among different socioeco-
nomic groups. Second, analysis of the proximate deter-
minants of the decline in the total fertility rate between
1970 and 1976 indicated fertility control by married women
to be the major factor involved; this conclusion was also
supported by evidence on the diffusion of contraception
among new regions and income groups. Third, accelerated
fertility decline coincided with a number of important
socioeconomic changes that could have had an impact on the
motivation to control fertility, including increases in
female educational attainment and possible aggravation of
inflationary pressures on the economic resources of low-
and middle-income households. The last two points sug-
gested a working hypothesis about the acceleration of
fertility decline: that it was triggered by the conver-
gence of two sets of forces--increased availability of
effective means of contraception and the emergence of
socioeconomic conditions conducive to smaller family
norms .
115
OCR for page 116
116
Because there is a lack of nationally representative
data combining information on contraceptive use and socio-
economic characteristics, the analysis in this chapter is
1 imited to the second set of forces. The pr inc~pal aues-
tion addressed is the extent to which the effects on
reproductive behavior of such modernizing forces as
increased household income and f emale educational attain-
ment (as measured by average parity) combined with other
factors affecting household behavior, particularly those
reflecting such major structural dimensions of the Brazil°
Ian economy as inflation and income distribution. Differs
ential patterns of change among rural and urban areas sug-
gested a division of ache chapter into separate sections
for these groups, though work on the latter was even more
rests i<:ted by data limitations .
URBAN WOMEN
As noted earlier, the acceleration of fertility decline
in Brazil coincided with a period during which lower- and
middle-income urban households were raising their consump-
tion expectations end beginning to realize them through
increased purchases of housing and other consumer dur
ables, including televisions and automobiles, with most
purchases made on the installment plan O Unequal treatment
of wages and credit obligations in Brazil's indexing sys-
tem made it more difficult for families to keep up with
payments, and even to purchase basic necessities during
periods of high inflation.
As was suggested earlier,
this, combined with increased knowledge of and access to
contraception, may have reduced family-size desires. This
explanation does not compete with a modernization frame-
work, but extends it to incorporate other structural
changes.
Along the research questions that need to be addressed
are the Following: (1) What measures in the available
data files can be used as appropriate indices of the mod-
ern~zing forces and economic pressures discussed above?
( 2) Mow should the relationship between these measures and
average par ity be specif led? Are the relationships line
ear? Should interactions be taken into account? (3) Can
the analysis be pushed beyond explanation of differentials
in average parity in 1970 and 1976 to an assessment of the
sources of change in fertility during that interval? In
other words, do declines in average parity reflect changes
in the composition of the population of married women
. . . ~
OCR for page 117
117
according to modernizing characteristics, or is it more a
case of changes in the parameters that ref lect the impact
of these var tables on par ity? The 1 atter are likely to
reflect structural changes, and one of the tasks here will
be to incorporate in the specif ication variables for which
such changes would be indicative of the specific struc-
tural forces hypothesized above, that is, increased eco-
nomic pressures on household resources.
It is important to recognize that questions In censuses
and large-scale surveys do not provide a great deal of
conceptual precision for the measurement of modernization
and its inf luence on fertility. Most of ache measures in
this chapter have appeared in the presentation of tabular
evidence in earlier chapters; these include income and
education, as well as age. A new variable that attempts
to measure households' relative economic positions has
also been added.
Table 38 lists the variables that have been selected
for the analysis of data on urban women. Variable labels
and a summary of var table def initions are shown. Average
parity (CEB), the dependent variable, is listed first.
In accounting for variation in this variable, the amount
of exposure to the risk of conception needs to be con-
trolled. This risk is associated with marital duration
Data on age at marriage are Provided
only in the 1976 data rile; maternal age is a less satis-
factory substitute, particularly at earlier ages when
there is greater variability because the marriage is
recent. To maintain comparability between 1970 and 1976,
regressions were run using maternal age (AGE) as a control
for exposure to risk. Women were first separated into
three broad age categories, with AGE used as a control
variable for each: (1) 20-24, (2) 25-34, and (3) 35-44.
To test the sensitivity of results to marital duration
(MDUR), the 1976 data were then run using the AGE control
variable.
The next two variables on the list relate to moderni-
zation. It has become fairly standard practice in analy-
tical approaches inf luenced by household economic theory
to use women' s average educational attainment (MED) and
their own and their husband's earnings (HINC) as variables
in analyzing differences In average parity (see Schultz,
1976). Women's earnings and educational attainment
reflect the opportunity costs of childrearing vis-a-vis
other uses of their time, while the earnings of husbands
and wives measure their available resources for child-
r ear ing and other activities. Theoretically, the effect
and maternal age.
OCR for page 118
118
TABLE 38 Var table Labels and Def ~ notions, Urban Women:
Braz il
Var table Def inition
CB
AGE
MED
HINC
MDUR
GAP
Average parity: number of live children ever born as
reported in 1970 census and 1976 PNAD survey.
Mother's age: in years.
Mother's education: natural logarithm of the number of
years of school completed, def ined as follows: MED-log
(years +1), so that log (0 years) "O.
Monthly earnings of the head of household: natural
logarithm of amount in 1970 cruzeiros.
Duration of marriage in number of years for currently
marr fed women .
Estimated log of head' s monthly earning (PINC) minus log
of observed earnings (ZINC).
PINC ~ p man, EXP, TAX), where :
For ~ years of school completed by head.
EXP ~ head ' a age years of school completed - 6 .
TAX - natural logarithm of value added tax per capita of
state in which woman resides.
On the number of children should be negative for cost
factors and positive for resource factors; in fact, how-
ever, both often turn out to be negative because increas-
ing income is usually accompanied by changing attitudes
about family size, including a preference for quality
rather than quantity of children.
Tabular evidence in earlier chapters indicated that
differences in average parity in Brazil negatively cor-
related with both income and education. However, the
question was raised of whether parity changes between
1970 and 1976 mainly reflected changes in the educational
and income levels of married women of reproductive age,
or whether other variables related to inflation and income
distribution also contributed to fertility decline. The
available census and survey data files provide only a few
threads of evidence on the latter. Increased ownership
of televisions, particularly among low-income households
OCR for page 119
119
likely to have purchased them on credit, suggests that
such factors could have been operative; increased female
labor force participation suggests that economic pressures
might have contr ibuted to the delay or termination of
ch ildbear ing .
However, exploratory attempts to incorporate television
ownership and female labor force participation in the
specif ication of the relationship between female educa-
tional attainment, husband's earnings, and average par ity
yielded unsatisfactory results. The problem was in iden-
tifying the endogenous effects of the employment and earn-
ings of women in reproductive ages on their fertility,
vis-a-vis the exogenous influence of those women's educa-
tion and their husbands ' earnings. The data f iles did not
provide other exogenous variables that could be used to
estimate fully identified parameters for female employment
and earnings. Consequently, the analysis was limited to
estimation of reduced-form coefficients; that is, only
exogenous variables were included on the r~ght-hand side
of the regression equation.
In view of this, and to permit continued pursuit of a
specification that would capture the effect of the rela-
tive economic position of a woman's household (as well as
possible changes in that position) on her fertility, an
approach based on the concept of relative income was
adopted. A households income is relative in that it may
be greater or less than the income stream that would be
expected on the basis of that household's human capital
endowments. A gap between observed and expected income,
if it existed, would indicate whether a household was more
or less vulnerable to outside economic pressures such as
inflation. The approach is consistent with Leibenstein's
( 1974 ) point about the relationship between income and
fertility: that while the overall relationship may be
negative, it may be positive within specific reference
groups, with higher fertility among higher-income house-
holds within a particular group.
This approach was operationalized by first estimating
the expected earnings of husbands using standard earnings
equations, and then determining what each husband's earn-
ings would be given his particular characteristics. The
estimates were based on husband' s earnings rather than
household income for s:mplicity's sake (estimates could
then be based on the characteristics of one individual
without consideration of household size and structure as
well as other income sources). Some of the limitations
of this approach are offset by the fact that the analysis
OCR for page 120
120
was restr ic ted to mare fed women with husbands present who
were living in single-family households.
Earnings are measured in logarithms of 1970 Cruzeiros
per month . The dif ference between repor ted (HINC) and
estimated (PINC) earnings provides a measure of the gap
described above. Variables in the earnings equation
include husband's education in years of schooling com-
pleted (HED); husband's experience (current age minus
years of schooling minus six [EXP1); and an index of the
level of industrial output in the state of residence,
which consists of the amount of value-added tax on manu-
f actur ing per capita in the state (TAX) . The latter
variable was added as a control for regional differences
in labor market conditions, which are known to vary widely
in Brazil. Earnings equations were estimated for 1970
and 1976, with the following results:
1970: PINC = 3.837 + .149 HEDUC
(85.3) (74.S)
+ .0008 EXP + .249 TAX
(1.13 (31.~)
1976: PINC = 4.013 + .158 aEnuc
(77.2) (79.0)
+ .011 EXP ~ .232 TAX
(10.9) (25.8)
R2 = .38
R2 = .37
Values of the t-statistic are shown In parentheses beneath
regression coefficients. A new variable (GAP) was defined
as the difference between PINC and HINC.
Other variables omitted from the analysis of indi-
vidual-level data studied in this chapter were infant and
child mortality. Mortality measures for use in analysis
of individual records can and have been der ived f rom these
data f iles (Metrics, 1981); however, the process of trans-
forming ratios of surviving children to children ever born
into a normally distr ibuted, jointly dependent var. table
would require controlling these ratios for length of expo-
sure to risk of mortality using parity progression ratios.
It was impossible to specify a statistically meaningful
causal relation between fertility and mortality in such
circumstances.
Unweighted means and standard deviations of the var$-
ables used in the regression analysis of differences in
average parity are presented in Table 39. The data are
broken down according to the three broad age categor ies
described above. Exploratory analysis of the relationship
between MEI) and CEB suggested a nonlinear specif ication.
OCR for page 121
i21
mABLE 39 Meansa (standard deviations) of Variables
Used in Analysis of Differences in Average Parity for
Urban Women, 1970 and 1976: Brazil
Age
Year and
Variable - 20-24 25-34 35-44
1970
CEB 1.78 3.47 5.00
(1.51) (2.43) (3.51)
MED 1.31 1.29 1.17
(0.79) (0. 83) (0. 84)
AGE 22.19 29 0 52 39.17
(1. 37) (2. 8S) (2. 84)
HINC 5.60 5.76 5.75
(0.92) (1.06) . (1.21)
GAP 0.11 -0.01 -0.04
(0. 72) (0. 81) (0. 97)
N 2,749 7,532 6,494
1_
CEB 1.45 2.89 4.61
(1. 21) ( 2. 01) t3. 11)
MID 1. 67 1. S9 1. 38
(0. 71) (0. 78) (a. 81)
AGE: 22.26 29.38 39 e29
(1. 37) (2.76) (2.76)
HINC 6.13 6.31 6.27
(1. 02) (1. 07) (1.11)
MI)UR 3. 20 8. 23 16.74
(2.26) (4.51) (5.94)
GAS 0. 09 -0 .04 -0. 02
(0.86) (0.15) (0.92)
N 2, 939 7, 691 6,491
aUnwe ighted sample means.
OCR for page 122
122
Consequently, MED was def ined as the logar i thm of a
woman's years of schooling completed plus one, so that
the logar ithm of zero years of schooling would equal
zero. As indicated above, HINC is also measured in
logar i thms .
S ince GAP is the res idual of the estimated value of
ZINC, it has a zero mean for the total study population,
but varies slightly from zero ~ n specif ic age categories.
Current earnings of husbands of younger women averaged
slightly higher than their estimated earnings' while the
average GAP for older women fell slightly below zero. I t
should also be noted that there are signif icant negative
correlations (.7 to .8) between HINC and t:;AP; this indi-
cates a tendency for current earnings to be lower than
expected earning among husbands with lower earning
levels, and for current earnings to exceed expected earns
ings by larger amounts among husbands with higher earn-
ings. A check of outliers in the earnings equations
indicated that these high zero order correlations were a
statistical artifact produced by cases at extremes calf the
income distr ibution: overestimation of earnings when
husbands repot ted zero earnings and underestimate ion when
they repot ted very high earnings. These correlations
disappeared when extreme cases were omitted; however, it
was decided not to exclude such cases from the analysis
cuff fertility differentials because the impact of the dif-
ferential between ZINC and PINC was of interest over the
entire range of the income distr ibution.
The data In Table 39 on average marital duration tM=R)
in 1976 suggest that analysis of fertility differentials
among younger women is likely to be quite sensitive to
differences in exposure to the r isk of conception because
of the selectivity within that age group toward women who
moor ~ -~ "arIV. The ever ace are at mar r iage for women aged
~ ~ _ _, ~ _ ,, , _ _ _ ~ ’ _ _
20-24 in 1976 was 19.1 years, `:o~arec1 ~o ADOS y="L. I.
women aged 25-29. Relative variation in MUIR is also
greater among younger women: the coefficient of variation
for that group was 71 percent compared to 55 percent for
25-34 and 35 percent for 35-44. Caution is suggested in
interpreting results for the youngest age category when
MOOR cannot be controlled.
Analysis of Urban CEB Differentials in 1970 and 1976
Results of ordinary least squares regression analyst s of
differences in CER for women grouped by broad age cate-
OCR for page 123
123
gories are presented in Tables 40 (1970) and 41 (1976)
Three alternative regression equations were estimated for
each data set. In each equation, the AGE var table was
included in an attempt to control for length of exposure
to r isk of conception within the broader age groups; an
additional equation substituting MDUR for AGE in the last
alternative is also shown for 1976 . Logar ithms of both
MED and HINC were employed af ter initial tests revealed
nonlinearity in the relation between these variables and
CEB. An education-income interaction term was also int-~o-
duced to determine whether the slope of CEB with respect
to income shifted with increases in education.
Interpretation of regression coefficients for MED and
HINC is facilitated by the logarithmic specification,
which makes them equally proportional to elasticities (the
percentage change in CEB for each one percent change in
MED or HINC) e Dividing either coefficient by the mean of
CEB gives the elasticity. Comparison between elasticities
calculated from the first regression equation for each of
the age groups in Tables 40 and 41 and averages for groups
of lower- and higher-income countries reported by Schultz
(1976) suggests that the responsiveness of BAR to changes
in MED and HINC among urban Brazilian women was at an
intermediate level, and that it moved in the direction of
higher-income countries from 1970 to 1976. Schultz found
that elasticities of average par ity with respect to both
mother's education and father' s earnings became increas-
ingly negative as the level of development increased.
This same tendency is observed in the Brazilian data
Elasticities for MED are greater (more negative) for
younger women, and increase f ram 1970 to 1976. The
results for HINC fall more clearly within the range of
Schultz's higher-income countries. A spry of the
results is as follows:
Age
Group MAD
1_
BINC
1_
1970
1_
20-24 -.30 -.38 -.08 -~07
25-34 -.22 -.31 -,07 -.05
35-44 -.20 -.23 -.07 -.09
Lower -.17 to -.06 +.05
(Schultz)
Higher -1.1 to -.19 -.11 to +.28
(Schultz)
.
OCR for page 124
124
He
o
m
L.
a)
ED
5:
o
_I
_I
N
\,.
~ m
Ld
U]
to
he
O
~ :~
~ he
O ~
Q
00
to
a: ~
A HE
~e
1 C)
C"
~ _
-
O~~ I~
0-
-
IY
41
S
Po
Yl
\0
· e
O ~
1
O
~4
e~e ~
_ ~r
_ 1
o
o
O
· .
O _
U~
o
o
'_4 0
· ~
~ _
~0
`0 ~ 0, 0. ~ ~ ~
_ ~ ~ ~ _ O _ ~ _ 01 _ ~ ~ ~ ~ \0
~4 C 4 C~ ~ ~ 1^ ~t ~ O ~ O ~ O t~
· 0 · · · e O ~ 0 ~4e 0 t~ ° ~ Oe ~ O ~e
_d _~ ~ ~ C~ ~ _ _ _
~ _
r _~
· e
0 r~
1 _I
-
_
_ 1
OCR for page 125
125
.-
o
o
.
,.
. ~
AS
.-
~ _
~ :
~ ^
: ~
~ .-
~ a.
~ ^
o
_
^
=.!
a
~ O
Z~
~ ^- ^ ~
. . . .
~ -= ~ O ~
. I .
~ ^ ~ ~ ~ ^
· . · . . · .
. ~ ~
e ~
:
:
. ~
0=
~ =
~ 2.
~ O ~
~ :D
~ a =
0
._
=
0
~ ==
~ O ~
~ e
e ~ ~
~ =.
OCR for page 134
134
by a combination of social, economic, and political
forces, many of them longstanding features of rural
Brazilian society. One of the most basic of these is the
combination of a limited supply of good land and increas-
ing population pressure; droughts and infertile soil have
plagued areas with a high concentration of rural popula-
tion, particularly in the Northeastern states . A second
force is the unequal distribution of land, with substan-
tial proportions of rural families being forced to eke a
subs~sten`:e out of minifundia holdings of fewer than f ive
hectares. A third is that there has been increased con-
solidation of land holdings ac:companyirtg the co~nmerciali
Nation of Brazilian agriculture; this process has received
additional stimulus from the energy crisis and Brazil's
resultant need to raise foreign exchange through agricul-
tural exports and increase production of alcohol as a sub-
stitute for petroleum imports. This lack of opportunity
in rural areas, combined with hopes of paid employment and
urban amenities, has motivated Brazil ' s rural-urban
migrations
While migration was the primary demographic response
to adverse socioeconomic conditions among the masses of
Brazil's rural population, other d~rahic processes were
affected, including fertility and family formation. Chap-
ter 3 suggested several hypotheses about how such changes
might be affecting rural fertility in Brazil. All focus
in one way or another on the role of children as both
immediate productive resources and longer-term ~nves~ents
for rural families, and on how changes in rural socioeco~
nomic and institutional conditions could have affected
this role e
One hypothesis relates to land availability. As the
amount of land available to farm families in core densely
settled areas is reduced, several forces that could moti-
vate not only out-migration, but also lower fertility are
set in motion. For families remaining in agriculture that
wish to transmit land to their children through inkier zL-
ance, large numbers of children will lead to an uneco-
nominal subdivision of plots; this can be avoided only by
having fewer children, forcing children to start their own
families later, Or encouraging the children' s migration.
The value of children as immediate productive resources
will also be affected to the extent that smaller plots
reduce the need for child labor.
The influence of these forces on reproductive behavior
is mediated by institutional f actors and by the availabil-
ity of new land in other areas. Research on land avail-
OCR for page 135
135
ability and fertility in rural Brazil using data from the
1970 censuses of population and agriculture found that
fertility was lower in more densely settled regions of
Southeastern Brazil and higher in areas of new rural set-
tlement in the Central-Western region (Merrick, 1974).
An extension of that research to Northeastern Brazil and
to the frontier settlements in the Amazon region did not
reveal similar patterns (Merrick, 1981). Several reasons
for this were suggested:
get underway until after 1970, and the very- unequal dis-
tribution of access to land in the Northeast was not as
conducive to the pattern observed in the South, here a
higher proportion of family farms were owner~operatede
Research with 1970 data also revealed the very severe
limitations of population census data for dealing with
links between rural demographic changes and socioeconomic
and institutional factors affecting those changes. These
limitations were overcome to some extent by combining pop-
ulation census data with information from the agricultural
census. Analysis of changes during the 1970s will have
to await the availability of detailed information from the
1980 censuses of agriculture and population. Onfortu-
nately, the PNAD surveys taken during the 1970s do not
provide the geographic detail required for linking them
to results of Me 197S agricultural census; moreover, the
PNAD sample does not include rural areas of the Central-
West and Amazon regions, so that the potential positive
effects on fertility of frontier settlement in those
regions during the 1970s cannot be examined.
Some features of the relationship between fertility
decline in rural Brazil during the 1970s and socioeconomic
change in the areas included in the PNAD surveys can be
studied. One of these relates to an institutional factor
mediating the response to increasing economic pressures,
such as scarcity of land and consolidation of land in
larger holdings, on Me demographic responses of farm
families. This factor is rural proletarianiiation, or the
shift from an owner~operator and farm family labor mode
of production to wage labor. As noted earlier, European
experience, as well as evidence from Southern Brazil,
indicates that lower fertility is more likely to result
from a scarcity-of land when farm families own their land
and reduced fertility allows them to maintain control of
the land. Rural proletarianization weakens this motiva-
tion. As noted in Chapter 3, Paiva (1982) disagrees with
this interpretation for Brazil. Be argues that prole-
tarianization reduces the value of children as farm
Amazon settlement really did not
OCR for page 136
136
laborers and provides incentives for reduced fertility,
including increased market work for both women and chil-
dren. The potential value of market work would be
increased by education, and by a shif t toward quality
rather than quantity of children.
The data with which to evaluate these two interpreta-
tions of Brazilian exper fence are scant. Tabular evidence
presented in Chapter 3 indicated that average parity was
indeed lower in reg ions with higher levels of proletar ian-
ization (measured according to the proportion of rural
household heads reported as employees ); irt the same
tables, however O average parity was higher rather than
lower for proletarian households within those regions.
While one must be caret ul to avoid the f allacy of compoo
sition, one must also be careful about biases ar ising f ram
the location of families when their characteristics have
For rural areas,
those data detect only that proportion of the proletar-
been measured in census and survey data.
ionized population remaining at the time News were
conducted; those who moved (either to urban areas or to
other rural areas) and who may well have had lower fer-
tility would not be included.
Another aspect of change in rural fertility relates to
the socioeconomic differentials observed in Chapter 2.
Bile these differentials were not as great as those
observed in urban areas, the data presented in Chapters 2
and 3 indicated that rural women with some education had
lower fertility than those with no education, and that the
proportion of rural women with some education increased
from 1970 to 1976. Also, rural women with no education
as a percent of all women aged 15-49 declined from about
21 percent to about 13 percent, a result of some comb~na-
tion of increases in ectuca~zona~ acca~rmer~c arcane Lucas
women, out-migration, and possible underrepresentation of
less-educated rural women in the PNAD survey. Such evi-
dence suggests that the role of education should not be
neglected in examining rural fertility differences.
Analysis of Rural CEO Dif ferentials in 1970 and 1976
As indicated above, neither the 1970 census nor the 1976
PNAD survey provides many meaningful measures of concepts
relevant to the analysis of changes in rural fertility in
Brazil. The multivariate analysis that follows is pre-
sented primarily to illustrate how the relationships
between differences in average parity and socioeconomic
OCR for page 137
137
vat tables descr ibed for urban women compare to the case
of rural women. No pretense is made of even approximating
a comprehensive explanation of changes in rural reproduc-
tive patterns based on these data .
Table 46 lists the variables included in the analysis.
Average parity (FOR) iS the dependent variable, and educa-
tion of women (MED) Is included in logarithmic form as it
was for urban women. Census and survey measures of income
and earnings had so little conceptual content for the
r ural population that they were quickly abandoned O lIuso
band' s education (=n) has been substituted as a proxy
measure of earnings potential, but no attempt was made to
calculate the gap between actual and potential earnings.
Two additional dummy variables were employed: one for
residents of Northeastern states (NE:) F included to pick
up the effects of regional differences in economic and
social structure not reflected in education, and one to
indicate if the household was proletarian {PROL), based
on whether or not He husband's status was reported as
employee or paid farm laborer.
Unweighted means and standard deviations of these
variables are shown in Table 47. Average CEB declined by
12 percent for married women aged 20-24, 6 percent for
TABLE 46 Variable Labels and Definitions, Rural Women:
Brazil
van table Def inition
-
C~R
AGE
MED
HEI)
NE
PROL
Average par itys namer of live children ever born
as reported in 1970 census and 1976 CHAD survey.
Moocher 's age: in yews.
Mother's educations natural logarithm of the nudger
of yews of school completed, defined a. follows
Clog (years +~), so that log (O years)-O.
Father's educations nat=^ logarithm of the number
of years of school completed, defined ~ follows
HED-log (years Al), so that log (0 ye~)-O.
Northe~ts day Variable equal to one for
residents of northeastern states.
Proletarians day variable equal to on. when
es~plo~nt stat" of father is rural wage labor.
OCR for page 138
138
TABLE 47 Meansa (standard deviations) of Variables
Used in Analysis of Differences in Average Parity
for Rural Women, 1910 and 1976: Brazil
Age Group
Year and
Variable 20-24 25-34
35-44
1970
1976
CEB 2.21 4.42 6.79
(1070) (2071) (3.82)
MED 0. ~2 0.53 0 0 41
(DO 66) (0. 67) (0. 61)
HED 0.53 0.56 0.49
(0. 66) (0. 67) (0. 64)
AGE 22.07 29.17 39.01
(1.42) (2.87) (2.79)
NE 0. 45 0.43 0. 43
{0. 49) t O. 50) (0. 50)
P - L 0.31 0.30 0.26
(0046) (0.46) (0. 44)
N 2,456 4,880 4,108
CEB 1.95 4.14 6.73
(~.50) (2.57) (3.62)
ME 0O90 0.78 0.61
(0. 75) (0 O 74) (0 O 70 3
lIED 0.84 0 0 76 0. 65
(0.74) (0.73) (0.71)
AGE 22 a 17 29 ~ 26 39 ~ 21
(lo42) (2~85) (2~89)
NE 0~32 0~33 Oc31
(Oe47) (0~47) (0~46)
PROL 0.48 0.43 0. 38
(0. 50) (0. 49) (0. 48)
N 2~344 5~247 4~615
al;lnweighted sample means.
OCR for page 139
139
those aged 25-34, and less than ~ percent for those aged
3 5-44. Relative var iation in average Rapt was somewhat
lower for rural than for urban women. For women in the
2 5-34 age group Of the coef f ic lent of var iation ( ratio of
the standard deviation to the mean) was 0.61 in 1970 com-
pared to 0.7 for urban women. In 1976, relative variation
increased, and was close to the level observed for urban
women. There was a greater proportional increase in aver-
age education for women than for men, with the greatest
increase occurring among ages 20-24. The Northeast was
represented less in 1976 because of a smaller sampling
fraction for that region in POND (as well as possible
underrepresentation in the Samoyed.
~__, The proportion of
women in proletarian households increased for all three
age groups, again with the greatest increase among younger
women.
Multivariate regression results for 1970 are presented
in Table 48, and those for 1976 in Table 49. Separate
regressions were run for each of the three broad age
categories, with a control for age within each category
as cell. Three regressions were selected for each group:
the first includes, in addition to AGE, the education
variables MED and ~ the second adds the dub variable
for the Northeastern states (NE); and the third includes
the dummy variable for proletarian households (PROLl.
Since both MED and "D enter in log form, their regression
coefficients are again proportional to elasticities
(which are derived by dividing the coefficients by the
sample mean value of I).
A sugary of the elasticities of If with respect to
average MED and BED is as follows:
Year 20-24 25-34 3S-44
.
MED ID MED BED ME ID
1910 -.16 -.06 -.10 -.02 -~04 -.01
1976 -. 16 - .10 ~ .15 -. 06 -.07 -. 08
Compared to the results for urban women, the responsive-
ness of CEB to increases in MED is one-third to one-half
as great for rural women, and the increase in responsive-
ness from 1970 to 1976 is smaller. Compared to results
for the range of countries reported by Schultz (1976),
results for older rural Brazilian women are close to the
bottom of the range of elasticities for lower-income coun-
tries (-.17 to -.04) and for younger women near the top.
Although addition of the dummy variable for residence
in Northeastern states (the second set of equations) does
not add significantly to the amount of variance explained,
OCR for page 140
140
~, _
-
m
C)
-
_
~4
d
. -
C)
o
~4
Q
L.
C.
~Q
·~4
U;
:- ·e
O
fo
~: ~
~ '
O ~
· ~o ~o
O
~ 3
E~: ·d
m
E~ x:
a
3'
z
8
o~
~ o
a: V
~ ~ ~ ~ ~ ~ U.
o o o o o o o o,
o o o o o o ~ ~ ~
o ~ o ~o o ~ o o o ~ o ~ o ~ o ~ o
~ ~ ~La _ ,_~ _ <7~~ _ ~~ _ ~ _ ~ _ ~ _ ~ _
a, ~o U1 ~ iD
~ ~ ~.
~ _ _ _ _ _
~' C'4 ~ _. ~ ~ ~ _ _
01 ~ ~ ~ O
O \0 0 \0 ~ `0 ~ ~ ~ ~ ~ ~ O ~ O ~ O
~ ° a O
O C-4 0 O4 0 C~ O ~ O \0 0 `0 0 ~ O ~ O ~
_~ _
_ _ _
O _ ~ _
~ ~ ~ ~ 0
0 ~ 0 ~ 0 r~
· a · 0 a a
O ~ O ~ O O
_ _ _
0 _ a~ ~ ~ ~ ~ _ o. _ ~ _
C~4 ~ C~ \0 ~ U' ~ ~ ~ ~ ~ `0
- 4 `0 ~ 1 ~ r~ ~ r~ ~o ~ ~o ~
· a 0 0 · 0 ° · · a a a
O _. O _' O ~ O ~ O U' O 1^
- - - - - -
a, _ ~ ~ r~ ~ ~ _ ~ _ ~ ~ ~4 ~ ~
0 ~ 0 ~ O ~ ~ u1 ~ r~ ~ ~ ~ ~ ~o
r~ ~ ~ ~ co ~ ~ ~ \0
° a ~ ~ a ~ ~ a
0 ~ 0 ~ 0 ~ 0 ~ 0 ~ 0 ~ 0 ~ O ~ O r~
_t _l _1 C~ ~ ~ _ _ _
_ _ _ _ _ _
r" _ ~o _ ~ _ c~ _ r~ _ ~ _ ~ _ \0 _ ~ _
0 \0 ~ ~ ~ ~ ~D O (D ~ ~ co ~ ~ (D u~ C u~
U1 ~ ~ ~ O ~ ~ ~ r~ 0 ~ ~4 0 0 ~ O r~
a ~ a · a · ~ a e a
0 c~4 0 r~4 0 r~ 0 ~ 0 ~ a ~ 0 ~ 0 0 0 0
I _ t I _ I _ I _ I _ I _ I _ I _
o
_ ~ ~ O O _ ~ _ ~ _ O _ ~ _ ~
_ _ . _ ~ ~ ~ S
N ~ ~ ~ ~ ~ q. ~ ~ - 1 0` O1 ~ C~ ~ ~
~ · · · e ~ ~ 0 0 ° a ~ · O 0 C
O ~ O ~ O ~t O ~ O `0 ~ 0 O4 0 ~ O N
1 1 1 1 1 1 1 1 1
a.
`:
4S
~ ~ ~ _ In _ r~ ~ ~ _ u' _ ~ _ ~ _ u~ _
4 '0 ~ ~ _~ _~ ~ ~ ~ r~ d t~ t~ ~ ~ ~ O
{m ~ t~ ~n ~@ ~ (~ ~o \n c~ ~ ~ ~ (D O ~ O
· a ~ a 0 ~ a ~ ~ ~ ~ ~ 0 ~ ~ ~ ~ ~
) ~ ~ ~ ~ ~ _1 ~ C~ ~ C~ O O O O O O
1 1 1 1 _1 1 ~ 1 _.
_ _ _
_ _ _ ~ _ ~ ~ _ _ _
~ ~ ~ ~ ~ ~ ~ ~ ~ ~ C~ ~
1 _ _ 1 _ _ 1 _ _ _
O - ·~~ U~
O1
as
-
o
z
OCR for page 141
141
CD
-
~:
s~
o
s~
~
-
c~ cn
a: x:
-
o.
O ~
~ Z
z
s"
. N
~ m
-
£
:~
o
a~
at
0
_
m
6 ~
~ ~:
o
o
o
· ~
~ _
In _ ~ _
~ U' ~ O4
r~ ~ r.
o ~ o ~
o ~ o ~
_ _
y
~,
~o
z
o
o.
:,
o
C'
o
o
O4 o
· .
r.
-
0 r"
r~ ~4
· ~
o U~
-
o
o
o
·
o _
-
o
O4 r~
· ~
O U1
-
o
o
o
· ~
~ _
U'
o O4
· ~
o ~
o
o
mo
· . ~
_ o
cr.
o ~
· ~
o ~
-
o ~
· ~
o ~
-
_ ~ _ r~ _ ~ _ ~o _ ~ ~ ~ _
~ o~ 0 ~ ~o ~r~ ~o t~~. ~ _~ ~ ~ ~ ~ ~ 0
r~ 0 ~ ~ ~ o, ~ ~ r~ ~ c.4 ~ r~ ~ ~ c*4 ~ r~
~ ~ e ~ · e · ~ · ~ ~ · · e ~ · ·
O ~ O ~ O ~ O \0 0 ~ O ~ O ~ O ~ O
1 1 1 1 1 1 1 1 1
_ ~ O _ O _ ~ _ ~ _ ~ _ ~ _ C~ ~
~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ O ~ O `° ~ 10 10 eC
·q ~ ~ t_ ~ r ~0 ~ 40 /X~ ~0 ~ 1^
~ ~ ~ O · ~ e ~ · ~ · ~ 0 ~ · ~ ·
O ~ O \0 0 \0 0 ~ O _. O _. O 10 0 It' O It, 11,
1 1 1 1 _. ~ _' 1 _1 1 1 1
_ _ _
~ _ ~ _
O O ~
~ _' 0 ~n
· 0 · ~
~0
1 ~ _
~ _
C~ ~
1 _
o
r~
I_ 1
r~
0 ~ 0 ~ r~ - 4
o ~ ~ ~ ~ ~
· · · · · ~
~ ~ o o o o
c)
~4
r~
o
~q :n
OCR for page 142
142
the regression coefficient is signif icant for the two
older age groups. This indicates an interaction between
education and residence rather than an additive effect.
Tests for the interaction suggested that MED slopes were
lower in the Nor the es t, though so few rural women in the t
region reported any education that it was difficult to
judge the signif icance of the results.
The third set of equations attempted to assess the
impact of proleSar ionization on average CEB. The results
were consistent with the tabular results in Chapter 3,
which suggested that CEB was slightly higher in praletar-
ian rural households c When other var tables were con-
trolled, however, this difference was signif icant only
f or women in the 25-34 age group in 1976. These results
conf irm the point made earlier that an adequate test of
the proletarian hypothesis will require a more precise
measurement of the process of proletar ionization that
captures present and previous res idence and occupational
status, as well as a r icher depiction of the institutional
forces underlying that process.
Changes in Rural CEB from 1970 to 1976
The change in average CEO from 1970 to 1976 that can be
accounted for in the regression analysis results from
increased educational attainment. In this connection, it
should be recognized that mos t of the observed change is
not explained by changes in the var. tables included in the
regressions. It Is hoped that the case study data
reported In Part lI of this report w~11 provide richer
insight into the nature of changes in rural fertility
than do aggregate census and sample survey data.
CONCLUSIONS
Most of the variance in average par ity that can be
explained by application of multi~rar late regression anal-
ys is to data on individual mar r fed women f rom the Braz il-
ian census and PNAD survey relates to modernization
variables--education and average earnings. Most of the
change that can be accounted for betweer~ 1970 and 1976
relates to increases in these two var tables. The attempt
to incorporate a var table measur ing the relative economic
position of urban households indicated that there was a
positive association between fer tility and relative eco-
OCR for page 143
143
nomic status; that is, CEB was higher on average for women
whose husbands' current earnings exceeded the level of
earnings that would be expected given their education and
other characteristics. However, decomposition of changes
in CEB from 1970 to 1976 did not show that a change in GAP
contributed to fertility decline.
These regression results do not suggest that increased
modernization was the only reason for the change in Bra-
zilian fertility. Changes in regression coeff icients and
in constant terms in the regression equations suggested
that a variety of unmeasured factors could account for the
unexplained variance. Measures that were available in the
census and survey data provided little insight into the
nature of such changes, whether related to increased
access to contraception through public or private chan-
nels, or to institutional changes associated with shifts
in the Brazilian raodel of socioeconomic development.
It should also be recognized that average parity is a
poor measure of fertility for purposes of accounting for
change. It is a cumulative variable, and comparisons of
averages for different groups at different dates measure
the result of a demographic process rather than the pro-
cess itself. As noted above, it would have been better
to use current fertility as measured by births reported
for the year prior to the interview' however, as reported
in Chapter 2, the performance of that measure inspired
more confidence in its robustness for work with subgroups
of the population. Attempts to derive other measures
(such as the length of the first open birth interval) from
these data files procured equally unrewarding.
Data limitations curtailed even further the analyst s
of differences and the decomposition of changes in the
average parity of rural women. The main finding for these
women was that increased educational attainment contri-
buted most to the explanation of variance, though only a
limited portion of the total variance was explained by the
regressions. Some doubts were raised about the hypothesis
that proletarianization contributed to fertility decline;
however, caution was suggested about the validity of using
data on rural women classed as proletarian at the time of
the interview to test this hypothesis, rather than using
the experience of the proletarianization process.
It was also impossible to link what the census and sur-
vey data reported about the demographic characteristics
of individual rural residents and their families to eco-
nomic characteristics of their farms and institutional
features of their localities. For this reason, the study
OCR for page 144
144
could not explore hypotheses relating fertility decline
to changes in land availability, reduced need for child
labor, and increased nonfarm economic activity for rural
women and children. The census and survey data f iles did
include questions on school attendance that mer feed fur-
ther study. Information on child labor was included In
agricultural censuses, but not In the demographic censuses
and surreys. If Brazil is ever to conduct a fertility
survey patterned after the World Fertility Survey, it
would focus ideally on these elements of rural socioeco-
nomic structure, as well as those aspects of change in
urban areas that are not covered by currently avail able
census and survey data.
Representative terms from entire chapter:
average parity
